Multifunctional self-assembly peptide recognition method and system

Abstract

The invention provides a multifunctional self-assembly peptide recognition method and system, and relates to the technical field of biological peptide recognition. The method comprises the following steps: firstly, obtaining peptide sequences of a self-assembled peptide and a plurality of biological functional peptides, and inputting the peptide sequences into an embedded layer of a polypeptide function prediction model for 21-dimensional vectorization; and learning combination modes and sequence features of amino acids through an amino acid association feature extraction layer of the LSTM network. Then, acquiring direct contact information among different amino acids on the basis of a fine-grained feature learning layer of a multi-head self-attention mechanism; and finally, designing a Transform multi-task classifier which comprises five sub-task classifiers which are used for respectively predicting the self-assembly function, the anti-cancer function, the anti-inflammatory function, the antibacterial function and the antioxidant function. According to the method, a multi-layer structure is adopted to carry out layer-by-layer deep learning and optimization on polypeptide sequence features, efficient recognition and classification of multifunctional biological peptides are achieved, and finally the model performance is comprehensively evaluated through the accuracy rate, the precision rate, the recall rate and the F1 value.

Description

Technical Field

[0001] The invention relates to the technical field of biological peptide recognition, in particular to a multifunctional self-assembling peptide recognition method and system. Background Art

[0002] Peptides are composed of short amino acid chains and have unique advantages such as specificity, low toxicity, and easy synthesis. According to domestic and foreign research reports, self-assembling peptides, antimicrobial peptides, anti-inflammatory peptides, antioxidant peptides, and anticancer peptides, especially multifunctional self-assembling peptides, have important application value in the field of biomedicine and have been widely used in tissue engineering, anti-infection, anti-inflammatory, anti-aging, and anti-cancer. However, the discovery and acquisition process of these peptides is complicated, and they face problems such as high experimental costs and difficulty in screening.

[0003] In the field of peptide functional recognition, accurately characterizing and analyzing the sequence features of peptide chains and their functional relationships is a key challenge in building effective recognition models. Peptides are composed of 20 amino acids, and their permutations and combinations directly determine the functional characteristics of peptides. Traditional recognition methods usually regard peptide sequence encoding as a sequence encoding problem and use language models (such as Transformer) for embedding representation. However, the direct application of language models faces two major limitations: on the one hand, the number of amino acid tokens is relatively small, making it difficult for high-complexity models to fully utilize the combination relationship between different tokens when capturing sequence features, and they can only rely on position encoding, which weakens the ability to characterize sequence details; on the other hand, the peptide chain input features only reflect sequence relationships and lack rich biochemical information. In addition, the amount of training data is limited, so Transformer is difficult to fully learn and characterize peptide features under small samples. Long short-term memory networks (LSTMs) have been proven to be effective in peptide sequence feature extraction and are suitable for relatively short peptide sequence lengths. However, it is difficult to fully extract the complex sequence features of peptides by relying solely on LSTM, especially in the absence of other biochemical information, and the model effect is limited. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a multifunctional self-assembling peptide identification method and system, aiming to improve the performance of the model in the identification of polypeptide multifunctionality and self-assembly function. At the same time, by introducing the functional activity of the polypeptide as weight information into the model optimization process, the joint loss function strategy is used for optimization, so that the model can still effectively capture and identify the multiple functions of the polypeptide under limited data.

[0005] To achieve the above object, the present invention adopts the following technical solution: a multifunctional self-assembling peptide recognition method, comprising the following steps: Step 1: Obtain the peptide sequences of self-assembling peptides, antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides and anticancer peptides, and vectorize the amino acid sequences using an embedding layer; Step 2: Extract amino acid correlation features based on LSTM network; Step 3: Based on the fine-grained feature learning layer of the Transformer multi-head self-attention mechanism, the hidden layer vector value calculated by each LSTM cell is used as the new feature representation of the amino acid for learning to obtain the direct and close connection information of different amino acids; Step 4: Multi-task classifier, including five subtasks of self-assembly, anti-cancer, anti-inflammatory, antibacterial, and antioxidant detection.

[0006] In a preferred embodiment, the length of the self-assembling peptide is ≤20.

[0007] In a preferred embodiment, the embedding layer uniformly represents amino acid sequences of different lengths as a 21-dimensional vector; In a preferred embodiment, the LSTM network effectively learns the combination mode and position order characteristics of different amino acids.

[0008] In a preferred embodiment, the Transformer multi-task classifier includes five sub-task classifiers, each of which is a multi-layer perceptron structure for outputting the classification probability of the corresponding task.

[0009] In a preferred embodiment, the performance of the model is comprehensively evaluated by accuracy ACC, precision P, recall R and F1 value.

[0010] The present invention also provides a multifunctional self-assembling peptide recognition system, which runs the above-mentioned multifunctional self-assembling peptide recognition method; it includes a preprocessing module, which is configured to: obtain the peptide sequences of self-assembling peptides, antibacterial peptides, antioxidant peptides, anti-inflammatory peptides and anti-cancer peptides, and use the embedding layer to uniformly represent amino acid sequences of different lengths as 21-dimensional vectors; the original feature and sequence feature extraction module is configured to: extract amino acid association features based on the LSTM network, and effectively learn the combination mode and position order features of different amino acids; the global sequence information extraction module is configured to: input the combined features into the fine-grained feature learning layer of the Transformer multi-head self-attention mechanism, and learn the hidden layer vector value calculated by each LSTM cell as a new feature representation of the amino acid to obtain the direct and close connection information of different amino acids and obtain the final amino acid representation features; the classification module is configured to: classify the peptide sequence based on the final amino acid representation features.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1) Embedding layer: A 21-dimensional vector is used to uniformly represent amino acid sequences of different lengths, solving the inconsistency problem caused by different sequence lengths.

[0012] 2) LSTM feature extraction layer: The LSTM network learns the amino acid combinations and sequential relationships layer by layer, effectively extracts the feature information of the sequence, and provides high-quality initial features for the Attention mechanism.

[0013] 3) Multi-head self-attention mechanism: By assigning different weight values ​​between positions in the sequence through the Attention mechanism, the model can enhance its recognition of complex internal associations in the sequence, thereby improving the feature extraction of peptide sequences.

[0014] 4) Multi-task classifier and joint loss: Multi-task classifier achieves comprehensive discrimination of multifunctional features through parallel processing. Joint loss uses resistance information as optimization weight, which can effectively improve the prediction accuracy of the model for multiple tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the ATT-LSTM network structure of a preferred embodiment of the present invention.

[0016] Figure 2 The figure is an overall flow chart of ATT-LSTM of a preferred embodiment of the present invention.

[0017] Figure 3 This is a schematic diagram of the comparative effect of ATT-LSTM of a preferred embodiment of the present invention on a self-assembling peptide (peptide length ≤ 20) test set with other methods.

[0018] Figure 4 This is a comparative effect table of ATT-LSTM of the preferred embodiment of the present invention and other methods on the antimicrobial peptide, antioxidant peptide, anti-inflammatory peptide and anticancer peptide test sets.

[0019] Figure 5 This is a diagram showing the effect of peptide sequence length on the self-assembly properties of ATT-LSTM in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0020] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0022] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0023] refer to Figure 1-5 In one embodiment of the present invention, a multifunctional self-assembling peptide identification method is disclosed. Based on the above technical innovation, this embodiment proposes a multifunctional self-assembling peptide prediction model (ATT-LSTM). The ATT-LSTM network model is a self-assembling, anti-cancer, anti-inflammatory, antibacterial, and antioxidant detection peptide identification algorithm based on multiple deep learning and with strong interpretability. Based on the proposed ATT-LSTM network model, the following processing is mainly performed on the peptide sequence to be processed: 1. Embedding layer quantifies amino acid sequences.

[0024] For a given peptide sequence, ATT-LSTM uniformly represents amino acid sequences of different lengths as a 21-dimensional vector.

[0025] 2. LSTM network.

[0026] This embodiment uses an LSTM network to extract the amino acid sequence features quantified by the embedding layer, including the combination mode of different amino acids and the position order features for effective learning.

[0027] 3. Multi-head self-attention mechanism.

[0028] The original features of amino acids encoded by the LSTM network are input into the fine-grained feature learning layer of the Transformer multi-head self-attention mechanism, and the hidden layer vector value calculated by each LSTM cell is used as the new feature representation of amino acids for learning to obtain the direct and close connection information of different amino acids, thereby capturing the global sequence information.

[0029] 4. Peptide sequence classifier.

[0030] The learned amino acid features are averaged and pooled to generate a representation of the peptide sequence, which is then fed into a multi-layer perceptron (MLP) for peptide activity classification. The multi-task classifier includes five subtasks: self-assembly, anti-cancer, anti-inflammatory, antibacterial, and antioxidant detection.

[0031] Furthermore, if Figure 1 and Figure 2 As shown, the multifunctional self-assembling peptide recognition method proposed in this embodiment may specifically include the following steps: The peptide sequences of self-assembling peptides (peptide length ≤ 20) and various biological functional peptides (antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides and anticancer peptides) were obtained and input into the embedding layer of the peptide function prediction model for 21-dimensional vectorization.

[0032] The amino acid association feature extraction layer of the LSTM network is used to learn the combination and sequence features of amino acids. Then, based on the fine-grained feature learning layer of the multi-head self-attention mechanism, the hidden layer output of the LSTM is used as the feature representation of amino acids to obtain the direct connection information between different amino acids.

[0033] Finally, a Transformer multi-task classifier was designed, which included five sub-task classifiers to predict self-assembly, anti-cancer, anti-inflammatory, antibacterial and antioxidant functions respectively, and the classification results were obtained.

[0034] 1. Evaluation plan This example mainly evaluates our model from the following two aspects: (1) Comparing this example with other algorithms on various evaluation indicators. (2) Testing the effect of peptide sequence length on the core technology proposed in this example, taking self-assembly characteristics as an example.

[0035] The recognition of multifunctional self-assembling peptides (peptide length ≤ 20) is actually a binary classification problem. Therefore, the present invention uses ACC (accuracy), Recall (recall rate), Precision (precision rate) and F1-score as evaluation indicators. The calculation of each indicator is as follows: (II) Experimental results (1) ATT-LSTM has great advantages over other advanced methods.

[0036] ATT-LSTM was trained and tested on five datasets: self-assembling peptides (peptide length ≤ 20), antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides, and anticancer peptides. TRN, LSTM, and GRU were compared on the self-assembling peptides (peptide length ≤ 20) test set. TRN, LSTM, GRU, and CNN-GRU were compared on the antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides, and anticancer peptides test sets.

[0037] The performance comparison results on the self-assembling peptide (peptide length ≤ 20) test set are as follows Figure 3As shown. Based on the comparison results, we found that ATT-LSTM showed the best performance on the self-assembling peptide (peptide length ≤ 20) test set, outperforming all other compared methods. ATT-LSTM has an accuracy of 0.9569, a recall of 0.9569, a precision of 0.9572, and an F1-score of 0.9570. ATT-LSTM is the only model with all evaluation indicators higher than 0.9, indicating that it has a strong ability to accurately identify self-assembling peptides (peptide length ≤ 20).

[0038] The performance comparison results on the test sets of antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides and anticancer peptides are shown in Figure 4 Based on the comparison results, we found that ATT-LSTM showed the best performance on the antimicrobial peptide, antioxidant peptide, anti-inflammatory peptide, and anticancer peptide test sets, outperforming all other compared methods.

[0039] On the anticancer peptide test set, ATT-LSTM achieved an accuracy of 0.7717, a recall of 0.7717, a precision of 0.7759, and an F1-score of 0.7710. ATT-LSTM is the only model with all evaluation indicators higher than 0.77, indicating its strong ability to accurately identify anticancer peptides.

[0040] On the antioxidant peptide test set, ATT-LSTM achieved an accuracy of 0.9040, a recall of 0.9040, a precision of 0.9049, and an F1-score of 0.9044. ATT-LSTM was the only model with all evaluation indicators above 0.90, indicating that it has a strong ability to accurately identify antioxidant peptides.

[0041] On the antimicrobial peptide test set, ATT-LSTM achieved an accuracy of 0.9415, a recall of 0.9415, a precision of 0.9411, and an F1-score of 0.9413. ATT-LSTM was the only model with all evaluation indicators higher than 0.94, indicating that it has a strong ability to accurately identify antimicrobial peptides.

[0042] On the anti-inflammatory peptide test set, ATT-LSTM achieved an accuracy of 0.7016, a recall of 0.7016, a precision of 0.7081, and an F1-score of 0.7037. ATT-LSTM is the only model with all evaluation indicators above 0.70, indicating its strong ability to accurately identify anti-inflammatory peptides.

[0043] In summary, ATT-LSTM significantly outperformed other state-of-the-art methods in the identification of self-assembling peptides (peptide length ≤ 20), antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides, and anticancer peptides. The high values ​​of ACC and P indicate that ATT-LSTM achieved high accuracy and consistency in the prediction of peptide functions.

[0044] In addition, given that the total number of peptides in the sequence space will grow exponentially as the length of the peptide increases, in order to further test the reliability of the ATT-LSTM model, taking the self-assembly characteristics as an example, the influence of the peptide sequence length on the ATT-LSTM model indicators was explored. 11 to 20 peptide data sets were selected to train the model, and 2 to 10 peptide data sets were used to test the model, exploring the effect of using different amounts of training data (1000 to 5000). Figure 5 As shown in the figure, as expected, the performance of the ATT-LSTM model improves with the increase of the size of the training dataset. It is worth noting that ATT-LSTM achieves a model index value of more than 0.85 with only 5000 training data in the 2-10 peptide sequence space, and is able to predict the complete sequence of oligopeptides regardless of the length of the peptide segment. This page demonstrates the effectiveness of ATT-LSTM in predicting the self-assembly properties of peptides with a huge sequence space.

[0045] Example 2 This embodiment discloses a multifunctional self-assembling peptide recognition system.

[0046] A multifunctional self-assembling peptide recognition system, comprising: The preprocessing module is configured to: obtain the peptide sequences of self-assembling peptides (peptide length ≤ 20), antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides, and anticancer peptides, and use the embedding layer to uniformly represent the amino acid sequences of different lengths as 21-dimensional vectors; The original feature and sequence feature extraction module is configured to: extract amino acid association features based on the LSTM network and effectively learn the combination mode and position order features of different amino acids; The global sequence information extraction module is configured to: input the combined features into the fine-grained feature learning layer of the Transformer multi-head self-attention mechanism, and learn the hidden layer vector value calculated by each LSTM cell as the new feature representation of the amino acid to obtain the direct and close connection information of different amino acids and obtain the final amino acid representation features; The classification module is configured to classify the peptide sequence based on the final amino acid representation features.

[0047] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0048] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A multifunctional self-assembling peptide recognition method, characterized in that: The following steps are involved: Step 1: Obtain the peptide sequences of self-assembling peptides, antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides and anticancer peptides, and vectorize the amino acid sequences using an embedding layer; Step 2: Extract amino acid correlation features based on LSTM network; Step 3: Based on the fine-grained feature learning layer of the Transformer multi-head self-attention mechanism, the hidden layer vector value calculated by each LSTM cell is used as the new feature representation of the amino acid to obtain the direct and close connection information of different amino acids; Step 4: Multi-task classifier, including five subtasks of self-assembly, anti-cancer, anti-inflammatory, antibacterial, and antioxidant detection.

2. A multifunctional self-assembling peptide recognition method according to claim 1, characterized in that: The length of the self-assembling peptide is ≤20.

3. A multifunctional self-assembling peptide recognition method according to claim 1, characterized in that: The embedding layer uniformly represents amino acid sequences of different lengths as 21-dimensional vectors; A multifunctional self-assembling peptide recognition method according to claim 1, characterized in that the LSTM network effectively learns the combination mode and position order characteristics of different amino acids.

4. A multifunctional self-assembling peptide recognition method according to claim 1, characterized in that: The Transformer multi-task classifier includes five sub-task classifiers, each of which is a multi-layer perceptron structure for outputting the classification probability of the corresponding task.

5. A multifunctional self-assembling peptide recognition method according to claim 1, characterized in that: The performance of the model is comprehensively evaluated by accuracy ACC, precision P, recall R and F1 value.

6. A multifunctional self-assembling peptide recognition system, characterized in that A multifunctional self-assembling peptide recognition method according to any one of claims 1 to 6 is performed; It includes a preprocessing module configured to: obtain peptide sequences of self-assembling peptides, antimicrobial peptides, antioxidant peptides, anti-inflammatory peptides and anticancer peptides, and use an embedding layer to uniformly represent amino acid sequences of different lengths as 21-dimensional vectors; The original feature and sequence feature extraction module is configured to extract amino acid association features based on the LSTM network, and effectively learn the combination mode and position order features of different amino acids; the global sequence information extraction module is configured to input the combined features into the fine-grained feature learning layer of the Transformer multi-head self-attention mechanism, and learn the hidden layer vector value calculated by each LSTM cell as the new feature representation of amino acids to obtain the direct and close connection information of different amino acids and obtain the final amino acid representation features; The classification module is configured to classify the peptide sequence based on the final amino acid representation features.

Images